ChatGPT Wallet takes 0% — but United banned its agents

OpenAI is taking a 0% cut on every purchase its agents make — and two days before the launch, United Airlines rewrote its Contract of Carriage to make sure those agents can never buy a ticket. The wallet solves payment. It does not solve permission, and permission is the whole ballgame.

What launched today

OpenAI announced ChatGPT Wallet on Friday, November 13. It lives inside dots, the always-on agents OpenAI introduced in September, and for now it is limited to US ChatGPT Pro subscribers, 18 and older, with eligible US-issued Visa, Mastercard, or American Express cards.

The mechanics are genuinely good engineering, and they are worth stating precisely, because the design is the story:

• You add a card once. Before each purchase you review the items, the total, and the card, then approve with a security check like Face ID or a fingerprint. Every approval creates a record of what you authorized.
• After approval, the payment network issues a token — a substitute card number with its own security code and expiry. The agent never sees your real card details. Neither does the merchant.
• Rewards and dispute protections carry over, because the charge still lands on your actual card. OpenAI built it with American Express, Mastercard, and Visa, and is working with Forter, Riskified, Sardine, and Signifyd on fraud detection and Adyen, Checkout.com, and Stripe on the payment side.
• It works with existing checkout pages. No Wallet-specific integration is required — the token just shows up as a card.
• OpenAI is not charging a commission on transactions at launch, per CNBC's reporting. The play is distribution, not fees.

1.2 billion people use ChatGPT every week, OpenAI says, and nearly 100 million of them pay, as of Q3. Many are already researching products there. The wallet closes the last step: research to purchase without leaving the chat. Eventually OpenAI wants full delegation — spending limits, categories, time windows, pre-approved purchases. At launch, every purchase still needs your explicit approval.

The part OpenAI can't ship

I opened the Commerce Partner Playbook OpenAI published alongside the wallet. It is live at developers.openai.com/commerce, and it is a six-step checklist for merchants: submit product feeds, check that your CDN or firewall doesn't block verified ChatGPT browser visits, accept Wallet payments on your existing site, build a plugin, add Sign in with ChatGPT, claim your business.

Step two is doing all the work. OpenAI's own announcement says it plainly: "websites need to ensure agents are not inadvertently blocked." Inadvertently is doing a lot of lifting there, because plenty of the blocking is deliberate.

Two days ago — November 11 — United Airlines announced it is restricting unauthorized third-party AI agents from its platforms, and revised its Contract of Carriage the same day to define AI agents and list agent bookings as a prohibited practice. Agents can still search fares. They cannot buy, modify, or service a ticket. Reuters reported the move on Wednesday; United cited credential exposure, unintended purchases, and MileagePlus miles spent without the traveler meaning it.

United is not the first. Amazon blocked Meta's AI agent Muse from buying on Amazon.com on September 20, and has barred other agents too. Walmart went the other way and partnered with Muse. Expedia and Booking are working with OpenAI. The industry is sorting itself into allowed and blocked, merchant by merchant.

Why this is the real test

Look at what OpenAI actually had to build here. The hard problems in agent payments were never the cryptography — tokens and biometric approval are solved technology, and OpenAI solved them correctly. The hard problem is that a purchase is a three-party transaction, and the third party gets a veto.

A token that the merchant's fraud system flags is a declined transaction. A checkout page that the merchant's firewall blocks is a dead end. A contract that defines your agent as a prohibited practice is a legal wall. OpenAI's playbook can ask merchants to configure their CDN to recognize signed requests via Web Bot Auth, and it can ask fraud vendors to recognize legitimate agent purchases. It cannot compel any of them.

That is why the 0% cut matters more than it looks. OpenAI is giving away the transaction economics to buy merchant adoption, because adoption is the only thing standing between this wallet and irrelevance. A wallet that works at every store that allows it is only as useful as the list of stores that allow it — and right now the biggest names are writing "no agents" into their contracts faster than OpenAI can publish playbooks.

The honest scoreboard for the next six months is not how many Pro users add a card. It is the allowed/blocked list: how many merchants leave the door open, and whether the ones that closed it — United, Amazon — ever reopen it. Payment was the easy half. OpenAI just shipped it. Permission is the hard half, and nobody at OpenAI can ship that.

Sources: OpenAI, "Introducing ChatGPT Wallet" (November 13, 2026); OpenAI Developers, Commerce Partner Playbook (developers.openai.com/commerce, fetched November 13, 2026); TechCrunch, "ChatGPT's new Wallet lets AI agents shop and check out for you" (November 13, 2026); Reuters, "OpenAI launches ChatGPT Wallet, letting AI agents pay for purchases" (November 13, 2026) and "United Airlines moves to block AI agents from its platform over security concerns" (November 11, 2026); CNBC reporting on the 0% launch fee and paying-subscriber count, via secondary coverage. The playbook pages are my own fetches; the United and Amazon blocks are as reported.